Off-campus UMass Amherst users: To download campus access dissertations, please use the following link to log into our proxy server with your UMass Amherst user name and password.

Non-UMass Amherst users: Please talk to your librarian about requesting this dissertation through interlibrary loan.

Dissertations that have an embargo placed on them will not be available to anyone until the embargo expires.

Author ORCID Identifier

0000-0003-2948-630X

Document Type

Open Access Dissertation

Degree Name

Doctor of Philosophy (PhD)

Degree Program

Public Health

Year Degree Awarded

2019

Month Degree Awarded

May

First Advisor

Nicholas G. Reich

Second Advisor

Leontine Alkema

Third Advisor

Kenneth P. Kleinman

Fourth Advisor

Laura B. Balzer

Subject Categories

Biostatistics

Abstract

Infectious diseases place an enormous burden on the people of the developing world and their governments. When, where, and how to allocate resources in order to slow the spread of a virus or deal with the aftermath of an outbreak is often the responsibility of local public health officials. In this thesis, we develop statistical methods for forecasting future incidence of infectious diseases and estimating the effects of interventions designed to reduce future incidence, bearing in mind the needs and concerns of those public health officials.

While most infectious disease forecasting models focus on short-term horizons (i.e. weeks or months), long-term forecasts made prior to the epidemic season may be more useful to public health officials. In Chapter 2, we make an annual forecasting model for dengue hemorrhagic fever incidence based on early season incidence, weather, and demographics. The predictions from this forecasting model outperform a baseline model based on the ten-year median on out-of-sample data. To our knowledge, this model makes accurate annual forecasts earlier in the year than any other dengue model on record.

After public health officials implement an intervention, whether a preventative action or a response to a developing outbreak, they may want to know whether that intervention was effective. In Chapter 3, we evaluate an effect estimation technique, called covariate-adjusted residuals, within a causal inference framework. This technique was originally developed for use in randomized trials, but has also been used in observational settings in ecology. Much research in the field of causal inference has focused on developing methods that account for confounding in non-randomized experiments. To our knowledge, we are the first to evaluate covariate-adjusted residuals from a causal inference perspective, and to develop an extension for use in observational studies.

In Chapter 4, we investigate whether using forecasts can improve the efficacy of effect estimation. In certain situations, forecasting can be used for covariate selection and dimension reduction that improves the performance of covariate-adjusted residuals in estimating the effect of an intervention. We used our findings to estimate whether an intervention for Zika reduced dengue hemorrhagic fever incidence in Thailand in 2016.

Included in

Biostatistics Commons

Share

COinS